Estructuras

To help understand the architecture we describe a basic usage scenario for the Social Cloud through a sequence diagram in Figure 6.2.

The process starts when a Facebook user discovers the Social Cloud for eResearch application. If they choose to add the application, permissions for the required user data are requested through Facebook.

Once the application has been added, the user is presented a form to determine their interests in various research areas – this is used to generate their interest signature. The interest signature generated is then compared against the project signatures of all the available projects to determine ap-propriate projects ordered (by projected interest) for this user. The user can then select any projects that appeal to them and the Social Cloud proceeds to create accounts for them at each of the various BOINC project servers on their behalf.

Suggestions are made to the user on the resource shares that they should

6.4. INTERACTIONS 51 allocate to the various projects that they have selected. These suggestions are based on the normalized resource share values of their friends. This allows for meaningful competition in the future.

The user is then prompted to install the BOINC client and provide cre-dentials to connect to the Social Cloud. The BOINC client pulls informa-tion regarding the projects that the user has selected along with their re-source shares from the Social Cloud. It connects to each of the project servers and downloads work units for processing. In due course, the re-sults of the processing are sent back to the project servers. Each project server verifies the results obtained and grants credits as appropriate.

The Social Cloud routinely queries credits for every user from individ-ual project servers. If a user is found to have achieved a credit milestone in a project, it is published to their Facebook wall. This is visible to friends and should generate cascades of interest. The user can also view the ap-plication at their convenience to check on their progress and that of their friends. They may also suggest the Social Cloud for Public eResearch to their friends to help drive its growth (and increase their social score).

In addition, the Social Cloud periodically processes data available to it to establish rankings and achievements for users based on the algorithms described previously.

User Facebook SocialCloud Project Server BOINC Client

Install BOINC client, provide SocialCloud credentials

Request user preferences

Figure 6.2: A simple example of interactions between all the actors associ-ated with the Social Cloud for Public eResearch.

Chapter 7

Results and Analysis

In this chapter, I will go through the analysis of the contributions of the Social Cloud of Public eResearch. I will first introduce a standard dataset that is used for social network analysis and explain how simulations on it have supported several of the assertions in this thesis. I will show how bringing together social networks and volunteer computing is beneficial to the cause of the latter. Those contributions that cannot be reliably studied through simulations, due to the inability to simulate human behaviour, are supported through logical analysis – only a large scale deployment would allow conclusive statements to be made. I also share results and conclusions drawn from a limited user study that looks at effectiveness of interest/project signatures and signature distances.

7.1 Visual Analytics Benchmark

In order to study the effects of the contributions of this thesis, I used a standard social network dataset to study and perform simulations that would generate reliable and reproducible results. The dataset is part of the Social Network and Geospatial benchmark [63] and is provided by the Human-Computer Interaction Lab at the University of Maryland. It was used for the IEEE Visual Analytics Science and Technology (VAST) 2009

53

Challenge.

The dataset consists of two tab-delimited tables – one of which de-scribes entities (people, cities and countries) and the other containing links between the entities.

The entities table consisted of 6016 entities with 6000 of them being persons, 12 of them being cities and the four remaining are countries.

Geospatial data being of little importance for the purposes of this thesis, the 16 non-person entities were omitted. The table of connections consists of 29,888 entries of which 12 were connections between cities and coun-tries. These 12 connections were similarly omitted.

The resultant dataset was further analysed and it was found that the minimum number of connections from any person in the set was 4. The maximum number of connections was found to be 449. The average num-ber of connections to any given person was found to be 9. Figure 7.1 shows the distribution of friend connections in the dataset. This distribution ex-hibits a very small number of users having a very large number of con-nections while the majority of the users have a small number of connec-tions. This distribution follows the Pareto principle or 80-20 rule [64] and is known to be representative of real world friend connections.

The graph shows that more than half of the people represented in the dataset have between 5 and 7 connections. This is a pessimistic model for simulation given that the average number of friends that a Facebook user has is around 120 [65]. So it is expected that results from real world social networks would significantly improve over the results presented in the sections following that use the VAST Challenge dataset.